✦ The Future BetsNarrow moat

Micron's four future bets look like four different projects — a new memory generation, a foundry partnership, a new module, and the largest construction program in its history. They are one bet worded four ways: that memory can stop being a commodity before the cycle comes back for these margins.

The near bet is HBM4, and it is already banked — every unit of 2026 capacity was sold out under multi-year agreements before the year began, with 2027 negotiations under way¹. The structural bet is HBM4E, where TSMC will fabricate the base logic die and each big customer can order its own customized version² — the first time in memory's history that the product is designed around the buyer rather than the standard. The sideways bet is SOCAMM2, a 256-gigabyte low-power module that tries to invent a server socket Micron can own³. And the decade bet is geography: roughly $200 billion of American fabs and R&D aimed at making about 40% of its DRAM in the United States⁴.

Notice what none of these is: a diversification. Lesser managements escape a commodity business by buying something else; Micron is spending every dollar deeper into memory, wagering that contracts, customization, new sockets and national borders can do for DRAM what consolidation alone never quite did — hold price when supply catches demand. The wager is being funded honestly, out of supercycle cash flow rather than debt, which is exactly when such bets should be made.

Grade them all on one future number: gross margin at the bottom of the NEXT memory downturn. The last trough produced a $5.8 billion annual loss; if the contracted HBM book, the custom dies and the new sockets are real moats rather than boom-time decorations, the next trough should stop far above zero. If margins round-trip to the old lows, these were not escape attempts — they were just what a peak looks like while it lasts.